[vorbis-dev] slightly different audio output

[Pattara Kiatisevi]

Thank you for all answers.

-I debuged the code and found that mdct_backward() returns quite different
values between float and int setting. (before MDCT, there error is almost
off-by-one, at most 1%).

-I'm using MDCT code from RC3 (not sure if the code in CVS is better?). I
have tried with 14-bit TRIGBITS and then modified the code to include
Segher's magic union:
http://www.xiph.org/archives/vorbis-dev/200202/0081.html. And run it at
TRIGBITS = 14, 20, 25, 30.

-the output from mdct_backward() seems to converge to a set of values when
increasing TRIGBITS but it doesn't converge to the result from floating
point version though.

-here is the result from decoders  1) floating point version 2) int with
30-bit mdct 3) int with 14-bit

rai16:~/cvs/oggonachip/software/vorbis/libvorbis-linux$ tail
audioout-float.txt
0290670 0262 0860 02ee 08a1 0377 08c5 03bc 0884
0290680 03c7 07ea 03be 073c 03d8 06c0 0459 0697
0290690 0506 0664 05bf 0606 06bd 05c6 07b6 0583
02906a0 0883 0528 097d 0505 0a71 04f9 0b23 04dc
02906b0 0beb 0502 0cc6 0564 0d6b 05aa 0df6 05eb
02906c0 0eb4 067a 0fad 0754 1072 07f8 10da 0846
02906d0 114e 08b1 11c8 093f 121c 09b8 128d 0a3d
02906e0 1318 0ac0 1396 0b20 141d 0b5e 1477 0b3c
02906f0 14a0 0aaf 14d1 09f1 14cb 08ed 14a4 07dd
0290700
rai16:~/cvs/oggonachip/software/vorbis/libvorbis-linux$ tail
audioout-int-30bit-mdct.txt
0290670 025e 085f 02ce 0896 034a 08b2 039d 087a
0290680 03ac 07df 03a0 073a 03de 06d7 0458 06a1
0290690 04f5 066b 05c2 061a 069e 05cc 0785 057b
02906a0 0865 052d 095e 050c 0a47 04f1 0b0e 04f1
02906b0 0bd2 0503 0c9e 054f 0d4a 059d 0dd6 05e8
02906c0 0e88 0672 0f79 0742 1046 07ec 10b1 083d
02906d0 1112 0896 1190 0928 11e5 09a5 1244 0a16
02906e0 12f9 0ab6 1395 0b16 13f5 0b37 144d 0b21
02906f0 149c 0ab9 14aa 09e7 150b 0939 14a5 07fd
0290700
rai16:~/cvs/oggonachip/software/vorbis/libvorbis-linux$ tail
audioout-int.txt
0290670 025e 085c 02cf 0895 0349 08b2 039e 087d
0290680 03ab 07df 03a0 0738 03de 06d6 0458 069f
0290690 04f2 066b 05c3 061a 069d 05cd 0784 057b
02906a0 0865 052e 095d 050e 0a46 04f2 0b0e 04f2
02906b0 0bd0 0505 0c9c 0552 0d48 059f 0dd7 05ec
02906c0 0e88 0675 0f77 0747 1045 07ec 10b0 083f
02906d0 1111 0895 1190 0927 11e4 09a8 1243 0a17
02906e0 12fb 0ab7 1395 0b18 13f6 0b3b 144e 0b22
02906f0 149c 0ab8 14a9 09e6 1509 093d 14a6 07ff
0290700

-has this int implementation of MDCT been proved to yield the same (or
almost the same) result with floating point version when using at high
TRIGBITS? Maybe my workaround is not correct..

Best regards,
Pattara
ps1.the output sound of this int version is ok, I can't differentiate..
ps2.any more information about how to do more frequency-domain analysis?

<p>
--
Please avoid sending me Word or PowerPoint attachments.
See http://www.fsf.org/philosophy/no-word-attachments.html
----------------------------------------------------------------------
Ott Pattara Kiatisevi 				   T L W G
M.Sc. INFOTECH Student, Stuttgart, Germany	http://linux.thai.net/
----------------------------------------------------------------------

On Tue, 9 Apr 2002, Segher Boessenkool wrote:

> > I thought you were talking about a difference of 1, not 1%.  Anyway, you're
> > correct; there ought to be a spec for tolerable variance, but I still
> > maintain it's a more complicated analysis.
>
> I didn't say it isn't.  I said to things:
>
> a) An error of more than 1u with 16 bit samples is unacceptable.
> b) The data in the original post has errors of more than 1% (much more,
> actually), which is much bigger than 1u.
>
> (1u is 1 smallest-representable-difference).
>
>
> Segher
>
>
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.
>
>

<p><p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.